Impact of AI-assisted chest radiography on clinical decision-making: A temporal analysis of CT referral rates and the proportion of positive CT in cardiology outpatient setting
This study demonstrates that the routine implementation of AI-assisted chest radiography in a cardiology outpatient setting was associated with a significant temporal increase in both chest CT referral rates and the proportion of referrals yielding positive findings, suggesting that AI aids cardiologists in identifying unexpected pulmonary abnormalities and optimizing diagnostic decision-making.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
In the quiet, often overlooked corners of a hospital, a routine chest X-ray is taken thousands of times a day. For a patient visiting a heart specialist, this image is primarily a map of the heart itself, showing its size, shape, and the health of the vessels that feed it. The doctor's eyes are trained to look for signs of heart failure or valve issues. However, the lungs sit right behind the heart, and they can hide problems that have nothing to do with the heart at all. A small spot that might be an early sign of lung cancer, or a patch of infection, can easily be missed when a doctor is focused on the heart. This is where artificial intelligence enters the story. These computer programs are designed to scan medical images and highlight areas that might need attention, acting like a second pair of eyes that never gets tired. The big question for doctors is not just whether the computer can find these hidden spots, but whether seeing them actually changes what the doctor decides to do next. Does the computer help the doctor catch more serious illnesses, or does it just create a flood of unnecessary alarms?
Researchers at Yongin Severance Hospital in South Korea set out to answer this question by watching how heart specialists used a specific artificial intelligence tool over a period of four years. They looked at nearly 9,300 adult patients who visited the cardiology clinic for the first time and had a chest X-ray taken. During this entire time, the hospital used a commercial software system that automatically analyzed every X-ray. The software did not just look for heart problems; it was also programmed to spot issues in the lungs, such as nodules, fluid buildup, or signs of infection. The system would attach a report to the X-ray image, highlighting any suspicious areas and giving them a score to indicate how likely it was that something was wrong. The researchers wanted to see if the presence of this digital assistant changed the doctors' behavior. Specifically, they tracked how often the heart doctors decided to order a more detailed scan called a chest computed tomography, or CT, and whether those CT scans actually found something that needed treatment.
The study divided the patients into two groups based on when they visited: an earlier period and a later period. As time went on and the doctors continued to use the artificial intelligence tool, the researchers noticed a clear shift in behavior. The rate at which heart doctors ordered CT scans increased significantly. In the first half of the study, about 1.6 percent of patients were sent for a CT scan. By the second half, that number had more than doubled to 3.3 percent. This was not just a random fluctuation; the data showed a steady, year-by-year rise in referrals. More importantly, the researchers found that this increase was not just about ordering more tests for no reason. The proportion of those CT scans that actually revealed a serious, treatable problem also went up. In the beginning, roughly 34 percent of the referred CT scans showed something significant, like an active infection or a suspicious mass. By the end of the study, that figure had climbed to nearly 47 percent.
This finding suggests that as the doctors gained more experience with the artificial intelligence tool, they became better at knowing when to trust its alerts. At first, the computer might have seemed like it was flagging too many things, but over time, the doctors learned to distinguish between minor, harmless findings and those that truly required further investigation. The study also noted that in many cases, the doctors ordered the CT scan before the official report from a human radiologist was even written. This highlights the speed at which the artificial intelligence tool provided information, allowing the heart specialist to make a decision immediately while the patient was still in the clinic. The researchers observed that the tool was particularly effective at finding unexpected lung problems in patients who had come in for heart issues, a group of doctors who might not otherwise be looking closely at the lungs.
The authors of the study are careful to note that they cannot say with absolute certainty that the artificial intelligence alone caused these changes. It is possible that other factors, such as the doctors simply becoming more experienced over the years or changes in the software itself, played a role. The software was updated during the study period, becoming better at detecting a wider range of lung issues, which may have contributed to the results. However, the data strongly suggests that the integration of this technology into the daily workflow helped the doctors identify more patients who needed further care. The study did not find evidence that the tool led to a flood of unnecessary tests that turned out to be harmless; instead, the tests ordered were increasingly likely to be the right ones.
Ultimately, this research paints a picture of a slow but steady adaptation. It shows that when artificial intelligence is introduced into a medical setting, it takes time for the human doctors to learn how to work with it effectively. The tool did not replace the doctor's judgment; rather, it seemed to sharpen it, helping the specialists focus their attention on the patients who needed it most. The study concludes that with continued use, artificial intelligence can become a valuable partner in clinical decision-making, helping to ensure that unexpected health issues are caught early, even in patients who are there for a completely different reason. The story here is not one of a machine taking over, but of a team learning to work together to improve patient care.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.